Bug #14714
closedRuby 2.5.1 Segmentation Fault in GC
Description
I am using Ruby 2.5.1, and running ruby on rails 5.1.4 with rspec-rails 3.7.2 and capybara 3.0.2.
When I run my feature tests, i.e. rspec spec/features, I get a segmentation fault. When I found the test in which the fault is occurring, and run that test by itself, then the fault does not appear. If I make the offending test pending,and run rspec spec/features again, then the fault still occurs but in another test. Accordingly, I cannot give you the exact code that is causing the fault.
I have attached the crash report log file.
Files
Updated by obromios (Chris Drane) over 6 years ago
I have reverted to ruby 2.4.4 and done rspec/spec/features and the problem does not occur, to this increases the probability that it is a ruby 2.5.1 problem.
Updated by nobu (Nobuyoshi Nakada) over 6 years ago
As it occurs in GC, it may reproduce with a few test by setting GC.stress
to true
.
Updated by renchap (Renaud Chaput) over 6 years ago
I think I am seeing a similar bug.
My test suite (Rails system tests, using Minitest and Capybara) is crashing on MacOS when using Ruby 2.5.1.
It is not happening with MRI 2.4.4, or on Linux (containers on Circle CI) using 2.5.1.
I first suspected a bug with ruby-vips
but as I can not reproduce it with 2.4.4, I think this is a bug in MRI (ruby-vips
issue for reference: https://github.com/jcupitt/ruby-vips/issues/174).
ruby-vips
is used to resize images using libvips
(via ffi). When I switch to another way of resizing (using MiniMagick — which shells out to ImageMagick), the segfault does not occur. I think it happens because when using ruby-vips
the resize is handled inside the MRI process and more memory is used.
The crash does not always happen, about 30% of the time the test suite finished. It sometimes crashes while resizing an image, but it sometimes also crashes later, in a test without any ruby-vips
calls. The stack trace is always the same, and related to GC.
I tried to reproduce the problem by isolating my image resizing code and calling it a few thousand times on the same images as my test, without success. Running my app full test suite on 2.5.1 on a mac is the only way to reproduce it.
I tried to enable GC.stress = true
at the start of the test suite but as the test suite uses a full-featured Rails app, launches a headless browser, ..., it is far too slow and I can not even get Rails to start.
Updated by renchap (Renaud Chaput) over 6 years ago
I also tried to isolate my vips-related code in a new Ruby program, and ran it 100 times with GC.stress = true
, without being able to reproduce.
Updated by obromios (Chris Drane) over 6 years ago
I tried putting GC.stress = true in the rails_helper file, and after 4.5 hours, no test has been completed, so this is too slow to be practicable for my machine.
I am not using ruby-vips.
Updated by renchap (Renaud Chaput) about 6 years ago
- ruby -v set to ruby 2.6.0preview2 (2018-05-31 trunk 63539) [x86_64-darwin18]
I just tested with 2.5.2 and 2.6.0-preview2 and I can still reproduce this when running my tests.
This still happens in GC:
-- C level backtrace information -------------------------------------------
0 ruby 0x00000001026a3ae7 rb_vm_bugreport + 135
1 ruby 0x000000010250cd43 rb_bug_context + 467
2 ruby 0x0000000102612951 sigsegv + 81
3 libsystem_platform.dylib 0x00007fff6caf7b3d _sigtramp + 29
4 ruby 0x00000001025274c3 rb_gc_mark_machine_stack + 99
5 ruby 0x000000010269275b rb_execution_context_mark + 171
6 ruby 0x00000001024f3ceb cont_mark + 27
7 ruby 0x0000000102531391 gc_start + 2529
8 ruby 0x00000001025306bc newobj_slowpath + 1324
9 ruby 0x0000000102530164 newobj_slowpath_wb_protected + 20
10 ruby 0x0000000102629e5d rb_str_resurrect + 61
11 ruby 0x000000010262dae3 rb_str_concat_literals + 195
12 ruby 0x0000000102684ebb vm_exec_core + 31675
13 ruby 0x0000000102691af8 vm_exec + 2600
14 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
15 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
16 ruby 0x000000010249b8a6 rb_ary_collect + 246
17 ruby 0x0000000102696e3b vm_call_cfunc + 299
18 ruby 0x000000010267ffc0 vm_exec_core + 11456
19 ruby 0x0000000102691af8 vm_exec + 2600
20 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
21 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
22 ruby 0x000000010249b8a6 rb_ary_collect + 246
23 ruby 0x0000000102696e3b vm_call_cfunc + 299
24 ruby 0x000000010267ffc0 vm_exec_core + 11456
25 ruby 0x0000000102691af8 vm_exec + 2600
26 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
27 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
28 ruby 0x000000010249b8a6 rb_ary_collect + 246
29 ruby 0x0000000102696e3b vm_call_cfunc + 299
30 ruby 0x000000010267ffc0 vm_exec_core + 11456
31 ruby 0x0000000102691af8 vm_exec + 2600
32 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
33 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
34 ruby 0x000000010249b8a6 rb_ary_collect + 246
35 ruby 0x0000000102696e3b vm_call_cfunc + 299
36 ruby 0x000000010267ffc0 vm_exec_core + 11456
37 ruby 0x0000000102691af8 vm_exec + 2600
38 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
39 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
40 ruby 0x000000010249b8a6 rb_ary_collect + 246
41 ruby 0x0000000102696e3b vm_call_cfunc + 299
42 ruby 0x000000010267ffc0 vm_exec_core + 11456
43 ruby 0x0000000102691af8 vm_exec + 2600
44 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
45 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
46 ruby 0x000000010249b8a6 rb_ary_collect + 246
47 ruby 0x0000000102696e3b vm_call_cfunc + 299
48 ruby 0x000000010267ffc0 vm_exec_core + 11456
49 ruby 0x0000000102691af8 vm_exec + 2600
50 ruby 0x000000010269f8ec invoke_iseq_block_from_c + 1084
51 ruby 0x0000000102697693 vm_call_bmethod + 435
52 ruby 0x0000000102697f02 vm_call_opt_send + 626
53 ruby 0x000000010267ffc0 vm_exec_core + 11456
54 ruby 0x0000000102691af8 vm_exec + 2600
55 ruby 0x000000010269f8ec invoke_iseq_block_from_c + 1084
56 ruby 0x0000000102697693 vm_call_bmethod + 435
57 ruby 0x0000000102680ed9 vm_exec_core + 15321
58 ruby 0x0000000102691af8 vm_exec + 2600
59 ruby 0x000000010269f8ec invoke_iseq_block_from_c + 1084
60 ruby 0x0000000102697693 vm_call_bmethod + 435
61 ruby 0x000000010267ffc0 vm_exec_core + 11456
62 ruby 0x0000000102691af8 vm_exec + 2600
63 ruby 0x000000010269f8ec invoke_iseq_block_from_c + 1084
64 ruby 0x0000000102697693 vm_call_bmethod + 435
65 ruby 0x0000000102697f02 vm_call_opt_send + 626
66 ruby 0x0000000102680ed9 vm_exec_core + 15321
67 ruby 0x0000000102691af8 vm_exec + 2600
68 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
69 ruby 0x000000010268be8e rb_yield + 158
70 ruby 0x0000000102495f49 rb_ary_each + 57
71 ruby 0x0000000102696e3b vm_call_cfunc + 299
72 ruby 0x000000010267ffc0 vm_exec_core + 11456
73 ruby 0x0000000102691af8 vm_exec + 2600
74 ruby 0x000000010269f26e invoke_block_from_c_bh + 398
75 ruby 0x000000010268c21e rb_yield_force_blockarg + 94
76 ruby 0x000000010249b8a6 rb_ary_collect + 246
77 ruby 0x0000000102696e3b vm_call_cfunc + 299
78 ruby 0x000000010267ffc0 vm_exec_core + 11456
79 ruby 0x0000000102691af8 vm_exec + 2600
80 ruby 0x000000010268fd03 rb_vm_invoke_proc + 451
81 ruby 0x00000001025ba8e2 rb_proc_call + 82
82 ruby 0x0000000102516813 rb_exec_end_proc + 355
83 ruby 0x0000000102516d95 ruby_finalize_0 + 133
84 ruby 0x0000000102516edb ruby_cleanup + 299
85 ruby 0x00000001025171af ruby_run_node + 63
86 ruby 0x0000000102493203 main + 99
Updated by renchap (Renaud Chaput) about 6 years ago
- Subject changed from Ruby 2.5.1 Segmentation Fault to Ruby 2.5.1 Segmentation Fault in GC
Updated by wanabe (_ wanabe) about 6 years ago
I think it is related to (or same as) https://bugs.ruby-lang.org/issues/14561.
Because (1) all SEGV is on macOS and (2) all crash points are in rb_execution_context_mark()
.
But this is merely speculation and weak conclusion.
Updated by renchap (Renaud Chaput) about 6 years ago
I tried to run my test suite (the only thing that I found triggering the crash) with FIBER_USE_NATIVE=1
and it worked on the first try, the crashed on the next one. Not sure they are the same bug, but probably related.
Updated by aselder (Andrew Selder) almost 6 years ago
I'm also running into this same SegFault while running a Rails app (https://bugs.ruby-lang.org/issues/15308).
Same symptoms, Mac OS X, and segfault in rb_execution_context_mark()
Updated by renchap (Renaud Chaput) almost 6 years ago
I just ran my test suite 5 times with Ruby 2.6.0 on macOS 10.14.2 and have not been able to reproduce this crash anymore.
Could you please all check if you still see this crash with the latest release?
Updated by aselder (Andrew Selder) almost 6 years ago
I believe this was fixed here: https://bugs.ruby-lang.org/issues/15362
Updated by wanabe (_ wanabe) over 5 years ago
- Related to Bug #14561: Consistent 2.5.0 seg fault in GC, related to accessing an enumerator in a thread added
Updated by wanabe (_ wanabe) over 5 years ago
- Related to Bug #15362: [PATCH] Avoid GCing dead stack after switching away from a fiber added
Updated by jeremyevans0 (Jeremy Evans) about 5 years ago
- Status changed from Open to Closed